The terms of answer engine optimization, defined in plain words, and exactly how aSERP measures each metric.
59 termsLast changed 7 Oct 2026
Find a term
Ask ChatGPT, Gemini or Perplexity where to buy a new TV and you get one written answer, not a page of links. It names some stores and leaves others out, and a new vocabulary has grown up around earning a place in that answer. This glossary defines that vocabulary in plain words. Where a term is also one of aSERP's metrics, the entry shows exactly how we compute it, what range it takes and what it cannot tell you, so the word on this page and the number on your dashboard always mean the same thing.
AI visibility is how often, how prominently and how accurately AI answers present a brand.
Why it matters. It is what answer engine optimization works toward. A brand can be named often but low in the list, or named and described wrongly, so no single number covers it.
In aSERP. aSERP reads it through five metrics, each per engine: Visibility, Prominence, Share of answer, Citation rate and Answer stability. Whether an answer describes you accurately, you check by reading what each engine answered on the Topics page; aSERP does not score accuracy.
An answer engine is an AI system that replies with a written answer instead of a list of links.
Why it matters. When a buyer asks an answer engine where to buy something, the reply names a few brands and leaves the rest out, so being named in that answer is a new place to be found. aSERP tracks six answer engines: ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek.
Also called: AI SEO, LLM optimization (LLMO), AI search optimization
Answer engine optimization (AEO) is the work that makes a brand more likely to be named, described accurately and cited when an AI system answers a question.
Why it matters. It is built on SEO, not a replacement for it. Google says that optimizing for its generative AI features "is optimizing for the search experience, and thus still SEO." What changes is the goal: a place in one written answer instead of a position in a list of links. GEO is another name for the same work, and has its own entry.
A brand mention is any appearance of a brand's name, or one of its other names, in an AI answer.
Why it matters. Mentions are the raw material of most of aSERP's metrics: Visibility counts the answers that mention a brand, and Share of answer weighs every brand an answer mentions.
In aSERP. aSERP reads every measured answer for the brand's name and the other names it goes by (its Also known as names), and counts a match under any of them as a mention of that brand.
Generative engine optimization (GEO) is another name for answer engine optimization, from a 2023 research paper; the two names are used interchangeably.
Why it matters. You will meet both names for the same work in tools, agencies and research. This site uses AEO throughout and mentions GEO here; the paper that coined the name is listed below.
Search engine optimization (SEO) is the work of helping pages get crawled, indexed and ranked by search engines, and it is still the foundation AEO builds on.
Why it matters. Google says that optimizing for its generative AI features "is optimizing for the search experience, and thus still SEO." Engines that search before they answer draw on search indexes, so a page that is not crawled and indexed is missing there too.
Share of voice is the industry's general term for a brand's share of all brand mentions across a set of AI answers, compared with its competitors.
Why it matters. Many tools report a share of voice, each with its own definition, so read how a tool counts before you compare its number with another tool's.
In aSERP. In aSERP the closest measure is Share of answer, which also weights prominence.
Not to be confused with.Share of answer: aSERP's measure: each answer's identified brands, weighted by how prominently each is placed, averaged over every measured answer.
A visibility audit is a one-off check of whether AI answers name a brand for a few questions.
Why it matters. A first look tells you whether to look closer. It cannot show a trend or a range, because answers change from one run to the next.
In aSERP. The free audit asks three questions on six engines, once. Its first-look score out of 100 is deliberately conservative and capped well below 100, because one capture cannot show a trend; the Visibility figure in a tracked workspace is the measure of record.
Not to be confused with.Visibility: a rate over many measured answers in a tracked workspace, with its confidence label; the audit's score is a capped first look from one capture.
In the product: the Answer stability card (How steady AI answers are)
Answer stability is a 0–100 score of how little the mix of brands aSERP identifies in AI answers changes from one measured day to the next; 100 means it did not change.
Why it matters. It tells you how much one day can tell you. When the mix of brands reshuffles every day, a single day says little; when it barely moves, the picture is steadier.
How aSERP measures it. It is 100 minus the Volatility Index, the average day-to-day change in that mix over the last 14 valid comparisons. It has no interval; its confidence label is rated on day-to-day comparisons instead.
Answer stability = 100 − Volatility Index
Not to be confused with.Presence steadiness: how steady one brand's own presence is, not the whole mix. Consistency: the word our home page and reports use for this kind of steadiness.
What it does not tell you. Whether you are named: the mix of brands can be steady without you in it.
When it shows a dash: the period does not have enough measured days to compare day to day.
An attributed citation is a citation aSERP links to a competitor through its websites or through how the answer named it.
Why it matters. It lets the competitor table show how often answers cite a rival, not only you.
In aSERP. On the competitor table of the Overview, your own row counts citations of your main domain, as your Citations card does. Each rival's row counts the citations aSERP attributed to that rival, which can include websites an answer named for them, and carries the label attributed.
In the product: the Citations card (Cited as a source)
Citation rate is, among the AI answers that name at least one website, the share that name or link the brand's own website.
Why it matters. An answer that names your website points its readers to your own pages. Citation rate shows how often that happens whenever answers name websites at all.
How aSERP measures it. aSERP counts a website only when the answer's own text names or links it; an engine's separate list of search results is not counted. The card's caption reads “Cited as a source”; in aSERP a source is always a website the answer's own text names or links. Only the brand's main domain counts (a leading "www." is ignored): if your main domain is example.com, citations to shop.example.com or to another domain you own are not counted. The interval uses the Wilson formula.
Citation rate = answers that name or link the brand's website ÷ answers that name any website
Not to be confused with.Citation: one website or page that an answer names or links; Citation rate is a share of answers. Visibility: an answer can name your brand without naming your website, and the reverse.
What it does not tell you. What the answer says about you. It also leaves out every answer that names no website at all.
When it shows a dash: no answers were measured in the period, or none of them named a website.
In the product:Answer stability; Presence steadiness on the Overview page
Consistency is the word our home page and reports use for how steady results are from day to day.
Why it matters. The same word has stood for two different measures, so this entry tells you which one you are reading.
In aSERP. On the dashboard today it appears as Answer stability, for the whole mix of brands, and as Presence steadiness, for one brand. The Answer stability card says it was previously called Consistency and that the figure is unchanged.
In the product:Presence steadiness on the Overview page
Presence steadiness is a 0–100 score of how steady one brand's own presence in AI answers is from one measured day to the next; 100 means identical every day.
Why it matters. It shows whether your own place in the answers holds from one day to the next, apart from how much the rest of the answer reshuffles.
How aSERP measures it. It compares one brand's presence on consecutive measured days. Days on which the brand is absent both times are not scored.
Not to be confused with.Answer stability: how steady the whole mix of brands is, not one brand.
What it does not tell you. How often you are named, or how prominently: read Visibility and Prominence for that.
When it shows a dash: the brand has no day-to-day comparison in the period, for example because it was absent on every measured day.
In the product: the Prominence card (How prominently you appear)
Prominence is a 0–100 score of how early and how strongly a brand is placed in the AI answers that name it.
Why it matters. Being named is not the same as being chosen. A brand given as the main answer is placed more strongly than one listed tenth or cited only as a source, and Prominence tells them apart.
How aSERP measures it. Per answer, aSERP weighs four parts: 50% the brand's role (the main answer 1.00, a list item 0.85, named in the text 0.60, cited only 0.25), 20% its list rank, 20% how early it is first named, and 10% repeats. The score is averaged over the answers that name the brand, with a bootstrap interval.
Prominence = 50% role + 20% list rank + 20% how early + 10% repeats, averaged over the answers that name the brand
Not to be confused with.Visibility: how often a brand is named at all; Prominence looks only at the answers that name it.
What it does not tell you. It is not a rank, and it says nothing about how often you are named. Read it beside Visibility.
When it shows a dash: no answers were measured in the period, or none of them named the brand.
In the product: the Share card (Your share of the answer); Answer share on the Competitors page
Share of answer is a brand's share of all the brands aSERP identifies in an AI answer, weighted by how prominently each is placed, averaged over every measured answer.
Why it matters. It shows how much of the answer goes to you rather than to your rivals: a brand named first and alone takes a larger share than one named last among several.
How aSERP measures it. For each measured answer, aSERP divides each identified brand's prominence by the total prominence of every brand it identifies in that answer; an answer that names no brand counts as 0 for everyone. The figure is the average over every measured answer in the period, with a bootstrap interval. The card's note reads “Of all brands named, vs competitors”; the number on it is weighted by prominence, as defined here.
Share of answer = the average, over measured answers, of the brand's prominence ÷ the total prominence of the brands identified in the answer
Not to be confused with.Share of voice: the industry's general term for a brand's share of all brand mentions; Share of answer also weights prominence. Visibility: how often you are named at all, whoever else is named.
What it does not tell you. Brands an answer names that aSERP cannot match to a known brand are left out of the total.
When it shows a dash: no answers were measured in the period.
Also called: mention rate, presence rateIn the product: the Visibility card (How often you show up); Brand visibility on the Competitors page
Visibility is the share of measured AI answers that name a brand.
Why it matters. Most other measures build on it: before an answer can recommend a brand or put it first, it has to name it. Engines answer the same question differently from one run to the next, so Visibility only means something as a rate over many answers, never as a single screenshot.
How aSERP measures it. In every scan, aSERP sends each tracked question to each tracked engine and reads every answer for the brand's name and the other names it goes by. For the period you choose, Visibility is the number of measured answers that name the brand divided by the number of measured answers. It is shown as a percentage, with a confidence label (High confidence, Directional or Low confidence) set from the number of answers behind it, the width of its 95% confidence interval and the share of the period's days measured. If an engine keeps failing to answer on a day, those missing answers are not counted as "not mentioned", and a period with no measured answers shows a dash, never 0%.
Visibility = answers that name the brand ÷ measured answers
Worked example (illustrative numbers)
Suppose a TV retailer tracks 10 questions, such as "Where should I buy a 65-inch TV online?", on six engines for 30 days, and 1,620 of the 1,800 planned answers are measured. If 405 of them name the retailer, its Visibility is 405 ÷ 1,620 = 25%. A rival named in 810 of the same answers has 50%. One answer can name both, so the two figures need not add up to 100%. The dashboard would show this with its confidence label; with only 10 questions behind it, treat a change of a few points as noise until it holds over more days.
Not to be confused with.Share of answer: your share of the brands aSERP identifies in each answer, weighted by how prominently each is placed. Citation rate: how often answers that name websites name yours; an answer can cite your website without naming your brand. Share of voice: the industry's general term for your share of all brand mentions.
What it does not tell you. Whether the answer describes you accurately, or recommends you. Read Prominence, and the answers themselves, for that.
When it shows a dash: no answers were measured in the period.
"AI unavailable" is the mark aSERP shows when an engine kept failing to answer on a day; those answers are not counted as "not mentioned".
Why it matters. An engine that could not answer is a gap in the measurement, not a sign that you were left out; counting it as "not mentioned" would make your Visibility look lower than it is.
A baseline is the level of a figure before a change, used for comparison.
Why it matters. Without a baseline there is nothing to compare the period after a change with, so the change cannot be read at all.
In aSERP. In aSERP's Experiments, the 14 days before the start form the baseline; a recommendation's baseline is captured the moment you choose Add to To-Do.
A comparison-eligible measurement is one taken the same way, on the day it reports, so it can be compared with another period.
Why it matters. A change between two periods only means something when both were measured the same way; otherwise the difference may come from the measurement, not from the answers.
In aSERP. aSERP declines a comparison it cannot make fairly, and says why: for example, “the method changed between the two periods” or “this period has too few measured days”.
A confidence interval is the range within which a figure's true value plausibly lies, given how much the measured answers vary; aSERP computes 95% intervals.
Why it matters. A figure without its range can look exact when it is not. Two figures whose intervals do not overlap are likely to differ for real; when either rests on few questions, wait until the difference holds over more measured days.
In aSERP. On the dashboard, Visibility and Citation rate use the Wilson formula, and Prominence and Share of answer a seeded bootstrap over answers, so the same data always gives the same interval. These intervals treat each answer as a separate observation. Answers to the same question tend to move together, so when a figure rests on few questions its real uncertainty is wider than these intervals show. aSERP's own research figures resample whole questions for that reason.
In the product:High confidence; Directional; Low confidence
A confidence label is aSERP's plain-words rating of how far to trust a figure: High confidence, Directional or Low confidence.
Why it matters. It folds the number of answers, the width of the range and the days measured into one plain word, so you know how much weight a figure can bear before you act on it.
In aSERP. For Visibility, Share of answer, Prominence and Citation rate, High confidence needs at least 100 answers behind the figure, a 95% interval no more than 10 points wide and at least 75% of the period's days measured; fewer than 30 answers is Low confidence. Below 5 answers a figure is too thin to rate, and its card shows Low confidence. The answers counted are all measured answers for Visibility and Share of answer, the answers that name the brand for Prominence, and the answers that name any website for Citation rate. Answer stability is rated on day-to-day comparisons instead: High confidence needs at least 14 comparisons, at least 75% of the period's days measured, and at least 70% of the possible day-to-day comparisons made; fewer than 7 is Low confidence, and below 5 it is not rated (shown as Low confidence). The headline above the cards shows High or Low; a figure its card calls Directional shows as Low there.
Correlation vs causation is the difference between two things moving together and one making the other happen; only a controlled test can show the second.
Why it matters. Most before-and-after readings in AEO are correlations. aSERP's Experiments say so in their own words: “Before/after comparisons are observational, not causal attribution.”
Coverage is how much of what was planned was actually measured.
Why it matters. A figure that rests on part of its planned answers can swing more than it should; coverage tells you how complete the picture is.
In aSERP. In the product it has two meanings. Under each card, a line says how many of the planned answers the figure is based on: that counts answers. The headline's Coverage reads Good or Building: that counts engines, and it is Good when at least two thirds of the tracked engines returned answers in the period.
An experiment is a planned change, such as a new page, followed by measuring the same questions before and after it.
Why it matters. Changing one thing and measuring the same questions before and after is the clearest way to see what moved; without a control group it shows what changed, not why.
In aSERP. In aSERP's Experiments, the 14 days before the start form the baseline, and every comparison is labelled observational, not causal.
In the product:informational; comparative; transactional
Intent is what the person asking wants: to learn, to compare or to buy.
Why it matters. A buyer asks different questions while learning, comparing and buying, so a question set that covers all three stages shows more of the decision.
In aSERP. In aSERP, every tracked question carries one of three intents: informational, comparative or transactional. The free audit asks one question of each kind.
A measured day is a calendar day (UTC) on which answers were actually captured; a figure belongs to the days it was measured.
Why it matters. Comparisons are fair only between periods measured the same way, so aSERP never fills in a missing day, and never presents an answer captured later as an earlier day's measurement.
Non-determinism is when the same question gets a different answer each time it is asked, even with the same settings.
Why it matters. It is why one answer is not a measurement. Even with the setting meant to make answers repeatable, 1,000 identical requests to one open model on a standard setup gave 80 different completions.
In aSERP. aSERP asks six answer engines (ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek) the same questions, the same way, every day, and reports each figure as a share of many answers with its range, never as one answer.
A snapshot is a single capture of AI answers at one moment: it shows what the engines said once, not how often they say it.
Why it matters. Answers change from one run to the next, so a single capture can make a brand look better or worse than it usually is.
In aSERP. The free visibility audit is a snapshot: three questions, six engines, captured once, and its result says so: “A snapshot, not a trend.” Tracked brands get a trend over many measured days instead.
A topic is a group of related questions about one subject, tracked and reported together.
Why it matters. Topics let you read results by buying decision rather than one question at a time, and the Topics page shows what each engine answered.
In aSERP. Changing the questions of a tracked topic, by rewording one or adding one, creates a new version of the topic, and some of its figures then count only the current version.
AI Mode is Google Search's conversational mode for longer questions and follow-ups.
Why it matters. It answers from the Google Search index, with query fan-out, so it matters to buyers who search on Google.
In aSERP. aSERP does not track it. To see your site's appearances in Google's AI features, use the Generative AI performance report in Google Search Console.
AI Overviews are Google's AI-written summaries shown above some search results.
Why it matters. Buyers meet them in Google Search, so they matter to your brand. They draw on the Google Search index, with query fan-out.
In aSERP. aSERP does not track them. To see your site's appearances in Google's AI features, use the Generative AI performance report in Google Search Console.
A hallucination is a confident statement in an AI answer that is false or invented.
Why it matters. Asked to identify the source of news excerpts, eight AI search tools gave wrong answers to more than 60% of 1,600 queries (Tow Center, March 2025). That is one reason to read what each engine says about you, not only whether it names you.
A large language model (LLM) is a model trained on very large amounts of text to read and write; it powers every answer engine.
Why it matters. What a model learned in training is one of the two routes into an answer, so what was written about you, widely and often, can shape what it says.
Model memory is what an AI model learned in training and can answer from without searching the web.
Why it matters. Asked without a search, a model answers from learned associations alone: it reflects what was written about you before its training data was gathered, mostly what was written widely and often. This route moves slowly, usually when the provider releases a new model. It is different from an assistant's memory of your earlier chats.
Query fan-out is when an engine splits one question into several related searches and combines what it finds.
Why it matters. A buyer's one question can become several searches, so a page can be found through a search the buyer never typed. Google calls running several related searches at once "query fan-out".
Retrieval-augmented generation (RAG) is a method in which a system first retrieves documents, then writes the answer from them.
Why it matters. It is the general pattern behind answers that cite sources: what is retrieved shapes what is written, though a page that is retrieved is not always cited.
A snippet is the short extract a search index returns about a page.
Why it matters. It can be all a model reads of your page. In one agency's study of 1,249 ChatGPT answers that searched the web, the model usually worked from a page's title, address and a snippet of roughly two hundred characters, so the first sentence under a clear heading may be all it sees of you.
The engines aSERP tracks are six answer engines: ChatGPT, Claude, Gemini, Perplexity, Grok and DeepSeek.
Why it matters. Engines disagree with each other more often than most teams expect, so a figure for all six together can hide where you are missing. Read each engine on its own.
In aSERP. Each engine has its own entry, under the same name, in the dashboard's Platform filter, so every metric can be read one engine at a time. aSERP does not track AI Overviews Not tracked by aSERP, AI Mode Not tracked by aSERP, Microsoft Copilot Not tracked by aSERP or Meta AI Not tracked by aSERP.
A web search tool is a capability an AI model can call to search the web while it writes an answer.
Why it matters. Engines that can search decide, question by question, whether the web will improve the answer; Google's documentation for Gemini says the model "determines if a Google Search can improve the answer." When an engine searches, a page can reach its answer much sooner than through a new model.
A citation is a website or page that an AI answer names or links in its own text.
Why it matters. A citation can send readers to a page, and it shows which websites an answer leaned on.
In aSERP. aSERP counts a citation only when it appears in the answer's own text; an engine's separate list of search results is not counted. So every citation figure in aSERP means a website named in the answer.
Not to be confused with.Citation rate: the share of answers naming websites that name yours. Brand mention: an answer can name your brand without naming your website.
Earned media is coverage a brand did not pay for or publish itself: press, independent reviews and reference sites.
Why it matters. It can feed both routes into an answer: models learn from what is written widely and often, and engines that search tend to draw on sources they already trust.
Freshness is how recently a page was published or meaningfully updated.
Why it matters. In one SEO-tool vendor's study, pages cited by AI assistants were about 25.7% fresher than pages in organic search results. A date changed without a real change is not freshness: show the real date, and change it only when the content changes.
In the product:Also known as in the Competitors section
An alias is another name or spelling a brand goes by, including one in another alphabet; answers that use it count as mentions.
Why it matters. As the free audit puts it, “Engines name you with whatever spelling they learned.” A name aSERP does not know is a mention it cannot count as yours.
In aSERP. You list a rival's aliases under Also known as, up to 10 for each rival, and the free audit asks for every name your brand goes by.
In the product:Competitors on the Brand Settings page
A competitor set is the list of rival brands tracked beside yours, so every figure compares you on the same answers.
Why it matters. Every comparison aSERP shows is against this set: a rival your buyers never consider only adds noise, and a missing one hides who is chosen instead.
In aSERP. Each rival in Brand Settings›Competitors shows Measured or Not measured yet. How many rivals you can track depends on your plan (see plans).
In the product:Entity Clarity on the Recommendations page
Entity clarity is how unambiguously engines recognise a brand and what it does.
Why it matters. An engine that confuses you with another brand, or cannot tell what you sell, cannot recommend you for the right question. One name, consistent facts and known spellings help on both routes into an answer.
An AI crawler is a program that fetches web pages for an AI company, to find pages that answers can cite or to collect text for models.
Why it matters. Blocking a search crawler removes your pages from that engine's search; allowing it does nothing on its own. A firewall or CDN rule can still turn away a crawler that robots.txt allows.
In aSERP. There are three kinds. Search crawlers find pages that answers can cite, such as OAI-SearchBot for ChatGPT, Claude-SearchBot for Claude and PerplexityBot for Perplexity. Training crawlers collect text to train models, such as GPTBot and ClaudeBot. User fetchers open a page when a person asks for it, and robots.txt may not stop them.
llms.txt is a proposed text file that summarises a site for AI tools; Google Search does not use it.
Why it matters. No measured effect has been found: an SEO-tool vendor's study of about 300,000 domains found no link between having the file and being cited. Publishing one is optional.
robots.txt is a file at the root of a site telling crawlers which paths they may fetch.
Why it matters. It is where you allow the search crawlers you want to be cited by, and decide about training crawlers separately. It is a request, not access control: user fetchers may ignore it, and a firewall can still block a crawler it allows.
Structured data is code that labels facts on a page, such as a price, a rating or an author, in a shared vocabulary such as schema.org.
Why it matters. An SEO-tool vendor's 2026 study compared 1,885 pages that added it with 4,000 matched pages and found no citation uplift, and Google says structured data "isn't required for generative AI search". Keep it accurate and matching the visible text, and never treat it as a ranking factor.
A first look: three of your buyers' questions, six engines, one capture. Daily tracking turns it into a trend with a range.
Your privacy choices
We use essential cookies to make this site work. With your permission we also use analytics cookies, to understand how the site is used, and marketing cookies, to measure our advertising. Nothing optional is switched on until you choose.
